Your SlideShare is downloading. ×
  • Like
  • Save
Metal artifact reduction
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×

Now you can save presentations on your phone or tablet

Available for both IPhone and Android

Text the download link to your phone

Standard text messaging rates apply

Metal artifact reduction

  • 1,666 views
Published

 

Published in Health & Medicine , Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
1,666
On SlideShare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
0
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Automated Segmentation and Interpolation in Sinograms for Metal Artifact Suppression in CT Wouter JH Veldkamp, PhD Raoul MS Joemai, BSc Aart J. van der Molen, MD Jacob Geleijns, PhD Department of Radiology, Leiden University Medical Center, The Netherlands
  • 2.
    • Metal implants such as prosthetic devices produce streak artifacts on the computed tomography (CT) images. The artifacts arise during the process of filtered back projection (FBP).
    Introduction
  • 3. Traditional image based approach
    • An early publication on metal artifact reduction was by Kalender et al. in 1987 1 . Most subsequent publications describe similar approaches. Their approach will be referred to as the ‘image based approach’.
    • Image based approach in short:
      • Segmentation of metal in reconstructed images (thresholding).
      • Forward projection of metal object to define metal region in original raw data.
      • Linear per view interpolation in metal region in original raw data.
      • Filtered backprojection to obtain an artifact reduced image.
    • Note that segmentation in reconstructed CT images is relatively easy since these images do not substantially suffer from over projection. Moreover, pixel values do not vary substantially within materials, facilitating segmentation in these kinds of images.
    • Note also that reprojecting regions to Radon space in the same geometry as the original raw data may be quite complicated, especially in the case of modern multi-row detector spiral data.
    1. W.A.Kalender, R.Hebel, and J.Ebersberger, "Reduction of Ct Artifacts Caused by Metallic Implants," Radiology 164, 576-577 (1987).
  • 4. Traditional image based approach For instance: W.A.Kalender, R.Hebel, and J.Ebersberger, "Reduction of Ct Artifacts Caused by Metallic Implants," Radiology 164, 576-577 (1987). original sinogram segmented sinogram interpol . sinogram back-projection uncorrected corrected Exclusion by interpolation segmented back-projection forward-projection
  • 5. Overview of the research: raw data based approach
    • The purpose of this work was to design an efficient and accurate metal segmentation in raw data to achieve artifact suppression and to improve CT image quality for patients with metal hip or shoulder prostheses. This approach is called: ‘ raw data based approach ’.
    • Advantages over the traditional image based approach:
    • Implementing segmentation and correction directly in original raw CT data or sinograms avoids the need for a complex Radon projection with the same geometry as the original raw data.
    • Moreover, potentially our method could be faster since the forward projection is omitted.
    • Furthermore, our method can also correct for metal objects that are completely or partly outside the reconstruction field of view since the disturbing metal parts should always be visible in the original raw data.
  • 6. Overview of the research
    • The raw data based approach incorporates two steps:
      • metal object segmentation in raw data (1) and replacement of the segmented region by new values (2) using an interpolation scheme, followed by addition of the scaled metal signal intensity.
    • Segmentation of metal directly in sinograms, makes the approach efficient and different from current image based methods. However simple segmentation methods fail in segmenting the complex metal signal in raw data 2 .
    • In this study segmentation is achieved by using a Markov Random Field Model (MRF).
    • Additionally, three interpolation methods were applied and investigated in this research.
    • To provide a proof of concept, CT data of five patients with metal implants were included in the study, as well as CT data of a PMMA phantom with Teflon, PVC, and titanium inserts.
    2. Kachelriess, M., Thesis/Dissertation, Friedrich-Alexander-Universität Erlangen-Nurnberg, 1998.
  • 7. Raw data based approach Original sinogram mask Interpol. sinogram Back-projection Back-projection uncorrected corrected High pass filter & thresholding MRF Canny filter Interpolation Selecting ROI
  • 8.
      • The iterative MRF 3-4 model is specified by giving the conditional probability distribution of a pixel label given its grey level and the labels of its neighbors.
      • We consider foreground (implant) and background as label.
      • A Gaussian model is used for representing the fluctuation of grey levels:
      • x i is a label of pixel i which can take two values: l =0 (background);1 (foreground).
      • α(l) is an offset value.
      • β(l) is a constant that models the a priori likelihood of labels to occur close to each other.
      • g(l) is the number of neighbors with class l.
      • The product of interaction parameter γ(l) and function h(C) models the interaction with edges in the MRF.
      • C represents the edge pattern (Canny filter output) within the current estimate of the foreground.
      • h(C) represents a low-pass filtered image of C to model edge interaction on a wider range.
      • y i represents the pixel value of pixel x i .
      • The parameters μ(l) and σ(l) model the statistical distribution of pixel values of class l.
    3. N.Karssemeijer, "A Stochastic-Model for Automated Detection of Calcifications in Digital Mammograms," Lecture Notes in Computer Science 511, 227-238 (1991). 4. W.J.H.Veldkamp and N.Karssemeijer, "Accurate seg mentation and contrast measurement of microcalcifications in mammograms: A phantom study," Medical Physics 25, 1102-1110 (1998). Raw data based approach
  • 9.
    • Patient with hip replacement. Fig 3a shows the first segmentation using high-pass filtering followed by an automatic thresholding technique 5 . Fig. b shows the ROI corresponding with Fig. 3a. The Markov random field uses the original data (Fig. 3b), the low-pass filtered output of a Canny filter (3c) and estimates of the mean local metal signal μ 1 (3d) and background signal μ 0 (3e). A first result of the MRF method is shown in Fig. 3f. After 5 iterations the final segmentation result is obtained (3g) and this result is used to correct raw data in the corresponding region (Fig. 3h).
    a b c d e f g h 5. T.W.Ridler and S.Calvard, "Picture Thresholding Using An Iterative Selection Method," Ieee Transactions on Systems Man and Cybernetics 8, 630-632 (1978). Mean foreground estimation: 2-D order-statistic filtering Mean background estimation: rough interpol. using mask (a) Raw data based approach
  • 10. The following empirically determined parameter values are used in the MRF model for segmentation of metal implants. Raw data based approach Parameter values 900 150 7 0 Background ( l=0 ) 300 0 0 0 Foreground ( l=1 ) δ γ β α
  • 11.
    • Interpolation within the metal region is performed in the original sinogram. Three interpolation techniques were investigated:
    • Linear interpolation is applied between the contour pixel sites at each viewing angle, indicated further as ‘per view interpolation’.
    • To possibly better preserve the structure of adjacent projections, pairs of pixel sites are determined on both sides of the implant based on shortest spatial distance (indicated further as ‘shortest distance interpolation’; based on 6 ). The result is smoothed using a 5x5 median filter.
    • A function is used that fills in pixels in the segmented implant region by smoothly interpolating from the pixels surrounding the region by solving Laplace's equation 7,8 , referred to as ‘smooth interpolation’.
    • Within the mask the difference between the original metal region pixel values and the corresponding interpolated pixel values is determined (as an estimation of the metal signal) and decreased to 10% of its value.
    • Subsequently this scaled difference is added to the interpolated pixel values within the mask. Thus the metal object remains recognizable after application of the metal artifact reduction process . 10% of the estimated metal signal is a pragmatic choice and appeared to give suitable results.
    6. M.Yazdia, L.Gingras, and L.Beaulieu, "An adaptive approach to metal artifact reduction in helical computed tomography for radiation therapy treatment planning: Experimental and clinical studies," International Journal of Radiation Oncology Biology Physics 62, 1224-1231 (2005). 7. J.H.Elder and R.M.Goldberg, "Image editing in the contour domain," Ieee Transactions on Pattern Analysis and Machine Intelligence 23, 291-296 (2001). 8. J.D.Wood and P.F.Fisher, "Assessing Interpolation Accuracy in Elevation Models," Ieee Computer Graphics and Applications 13, 48-56 (1993). Interpolation and signal addition
  • 12. Overview of interpolation approaches
    • Three approaches for interpolation in raw data
      • Per view
      • Shortest distance
      • Smooth
    Detector elements Viewing angle
  • 13. Signal addition a pragmatic method was applied Original signal Interpolated signal Interpolated signal + fraction of original signal
  • 14. Phantom
    • A Polymethyl methacrylate (PMMA) phantom used in this study.
    • The phantom has dimensions of 15 cm depth x 32 cm diameter.
    • The phantom contains 9 holes with a diameter of 1.5 cm. The inserts were constructed from different materials: Teflon simulating bone, Polyvinyl chloride (PVC) simulating fat tissue, and titanium simulating metal prostheses.
    • Two spiral acquisitions of the phantom were performed: one with titanium insert and one without titanium insert (a PMMA insert was used instead).
    • Data acquisition parameters: beam collimation 64 x 0.5 mm, tube voltage 120 kV, tube current 350 mA, scan field of view 400 mm.
    • All images were reconstructed at 1 mm slice thickness and 1 mm reconstruction interval.
  • 15. a b c d e 3 1 2 4 5 Metal artifact reduction in phantom images. Fig. a shows the uncorrected scan and the ROIs. Among other things the effect of different interpolation methods is shown. Fig. b shows a detail of the corrected image using the per view interpolation, Fig. c shows the same detail corresponding to the smooth interpolation and Fig. d corresponds to the shortest distance interpolation method. Finally, in Fig. e a result without adding the scaled projections to the interpolated values is shown (using smooth interpolation). Note that the streaking pattern is no longer evident in the images, but the region of reduced density between the objects is still present in the images and more evident in Fig. b (per view interpolation). Results
  • 16. Results Mean CT number and the standard deviation, both measured in HU, for pixels belonging to different inserts (ROIs) in the phantom are determined as a measure of distortion. Values are based on measurements in 40 consecutive reconstructed slices. 88 +/- 56 117 +/- 33 -53 +/- 27 -52 +/- 36 862 +/- 48 Smooth interpolation without signal 90 +/- 56 119 +/- 34 -54 +/- 28 -52 +/- 36 865 +/- 48 Smooth interpolation 99 +/- 51 118 +/- 34 -52 +/- 28 -53 +/- 37 867 +/- 48 Shortest distance interpolation 76 +/- 57 118 +/- 34 -67 +/- 28 -49 +/- 36 841 +/- 47 Per view interpolation 101 +/- 98 116 +/- 72 -64 +/- 85 -52 +/- 63 889 +/- 82 No artifact suppression With titanium 117 +/- 36 117 +/- 36 -62 +/- 30 -59 +/- 36 902 +/- 49 Without titanium Original raw data 5: PMMA 4: PMMA 3: PVC 2: PVC 1: Teflon Phantom configuration / Correction method Raw data type Region of interest
  • 17. Patients
    • Patient data
    • Spiral CT scans of 5 patients with unilateral prosthesis were acquired:
      • 4 patients have shoulder prostheses and one patient has a hip prosthesis.
    • Data acquisition was performed using the following parameters:
      • beam collimation 64 x 0.5 mm, tube voltage 120-135 kV, variable tube current (automatic exposure control was used), scan field of view 400-500 mm.
      • All images were reconstructed at 1 mm slice
      • thickness and 1 mm reconstruction interval. The images were reconstructed with a soft convolution kernel (FC12).
  • 18. Results Patient with hip implant. At the left slices of the original scan data are shown. At the right corresponding slices are shown that correspond to corrected raw data (smooth interpolation). The window center and window width were respectively 150 and 700 HU for all slices. Four different slices are shown consecutively.
  • 19. Results Results of four other patients with different implants. At left original slices are shown and at the right slices from corrected raw data are shown (smooth interpolation). For left and right images identical window center and window width was chosen.
  • 20. Results Results of four other patients with different implants. At left original slices are shown and at the right slices from corrected raw data are shown (smooth interpolation). For left and right images identical window center and window width was chosen.
  • 21. Results Results of four other patients with different implants. At left original slices are shown and at the right slices from corrected raw data are shown (smooth interpolation). For left and right images identical window center and window width was chosen.
  • 22. Results Results of four other patients with different implants. At left original slices are shown and at the right slices from corrected raw data are shown (smooth interpolation). For left and right images identical window center and window width was chosen.
  • 23. Results
    • For all patients, a fellowship-trained radiologist judged original and artifact suppressed images for visibility and assessment of relevant anatomical details and structures.
    • For artifact suppressed images, the raw data based method using smooth interpolation was involved. The radiologist quantitatively judged the images (original and processed) on a score from 1-5 (1 = bad image quality, 2 = moderate image quality, 3 = sufficient image quality, 4 = good image quality, and 5 = excellent image quality) for each of the 5 patients.
    • The average score for original and processed images was respectively 2.2 and 3.8 .
    3.2 2.0 Assessment of large anatomical structures 4.0 3.2 Neighboring bone loss 4.0 2.4 Prosthesis Assessment of 3.8 1.8 Small vasc. str. 4.0 1.8 Lymph nodes Visibility of small anatomical structures Supressed Original Average score Visibility and assessment
  • 24.
      • We have described a method to reduce the magnitude of CT artifacts due to metal prostheses. In both phantom and patient studies, this approach resulted in substantial artifact reduction.
      • We developed a segmentation method that is capable of segmenting metal structures in original raw CT data. The method is based on the use of Bayesian techniques and application of a Markov random field model.
      • Artifacts in CT data of a phantom and five patients were fully automatically suppressed. The general visibility of structures clearly improved. In phantom images, the technique showed reduced SD close to the SD for the case where titanium was not inserted, indicating improved image quality. HU values in corrected images were different from normal values for all interpolation methods. Subtle differences between interpolation methods were found.
      • The interpolation methods appeared to give similar results but small differences may exist: the smooth interpolation method and the shortest distance method appeared to give more consistent results in the phantom study than the per view interpolation method.
    Conclusions en discussion
  • 25. Conclusions en discussion
      • Our method is different from other papers on this subject. Most methods so far describe segmentation in reconstructed images and use a forward projection to replace projections in raw data.
      • Implementing segmentation and correction directly in original raw CT data or sinograms avoids the need for a complex Radon projection with the same geometry as the original raw data. Moreover, potentially our method could be faster since the forward projection is omitted. Furthermore, our method can also correct for metal objects that are completely or partly outside the reconstruction field of view since the disturbing metal parts should always be visible in the original raw data.
      • In summary, the Markov random field based segmentation method in raw data in combination with a relatively simple interpolation method (i.e. smooth, per view, or shortest distance interpolation) allows for a significant improvement of images that are corrupted by metal artifacts.
  • 26. THE END For further information contact: